Skip to main content

Tesla AI Day Coverage

This is full coverage of Tesla's AI Day and everything that was seen or talked about that was of note.

Tesla AI Day is currently 45 minutes late. At the moment, we have a full self-driving video happening showing off Tesla's autonomous driving, while listening to techno and dance music.

6:48 PM MST

Elon is on the stage now talking about AI Day and a brief overview of their full self-driving. He has passed the mic and time over to Andrej Karpathy, Tesla's lead in AI Development.

Talking About FSD

Andrej jumps right into the technical details of the neural network showing 8 cameras and Tesla's view of the road from the neural network's standpoint.

There are 3 learning HydaNets: Object Detection, Traffic Lights, Lane Prediction

Andrej mentioned that they first tried doing object detection with single images from the 8 cameras on the Tesla vehicle, but that wasn't working because objects could span multiple cameras. The 8 cameras needed to be fused as one view.

Andrej talked about how the neural net predicts what is ahead based on what it is currently seeing. This is what a human does when they see the road.

A big part of what Tesla does is mapping the area in real-time, whereas most other auto makers are using predefined maps. The predefined maps won't work long-term because the environment can change.

A member of the AI team was talking about how a car makes a lane change to the left in order to prepare for a left turn. In about 1.5 seconds, thousands of scenarios of that exact maneuver are analyzed and the car chooses the most optimal one. Incredible.

Tesla will not only be taking data from real world situations, it will also take that data and create a simulation like a game, which it will also use to train the FSD software. Wow.

Tesla has a number of tools to help its developers make changes to the neural network. I imagine that there's going to be a lot of checks and balances in place.

Talking About Dojo

The goal of Dojo is to achieve the best AI training performance and enable larger and more complex nodes.

Tesla has designed its own chips and Dojo is on par with the super computers of the world and it is still not enough. They will still be building it to be more powerful. There are fast chips connected to other systems, connected more and more...

Tesla will continue to refine and upgrade its Dojo hardware. Elon expects it to be operational next year.

Tesla Robots

Tesla unveiled a humanoid-like robot and they want people to help work on it. It's goal is to be friendly and to do repetitive and mundane tasks.

What did you think of Tesla AI Day? Was it too technical? Is Tesla further ahead than we think?

Leave your comments below, share the article with friends and tweet it out to your followers.

Jeremy Johnson is a Tesla investor and supporter. He first invested in Tesla in 2017 after years of following Elon Musk and admiring his work ethic and intelligence. Since then, he's become a Tesla bull, covering anything about Tesla he can find, while also dabbling in other electric vehicle companies. Jeremy covers Tesla developments at Torque News. You can follow him on Twitter, Facebook, LinkedIn and Instagram to stay in touch and follow his Tesla news coverage on Torque News.